LoginSign Up
← Back to all sparks
O

Ollama

AI-ASSISTANTS
Velocity6.3

Run and manage large language models locally on your own machine.

ollama.com ↗

Ollama v0.40.16 adds the 65B Ripple Preview coding model and cuts Qwen3.8 Flash memory on Macs.

local-llmcoding-agentsagent-hostsopen-modelsapple-silicon
◆Current state
Ollama keeps a weekly rhythm of model additions plus agent-host plumbing. v0.40.16 adds Ripple Preview, a 65B dense model for agentic coding, cuts Qwen3.8 Flash Next memory use on Apple silicon by about 27 GiB, makes `ollama launch claude` always route Claude Code to the selected model, and stops /v1/responses from silently truncating. A v0.40.17 candidate has started with an MLX version bump.
◆Where it's heading
Each release pairs a new open coding model with fixes for how Ollama behaves behind other agents, from Claude Code routing to OpenAI-compatible endpoints. The MLX runner keeps getting attention, which points at Apple silicon as a priority platform.
◆Prediction
Expect v0.40.17 within a week with another model addition and more MLX runner work.

◆Recent moves

  1. 3d ago

    v0.40.17-rc0: MLX version bump

    A release candidate that only bumps MLX and MLX-C. Engine upkeep at the start of the next cycle.

    View source ↗
  2. 4d ago

    v0.40.16: Ripple Preview model, 27 GiB less memory for Qwen3.8 Flash

    The stable cut adds Ripple Preview, a 65B dense agentic coding model, trims Qwen3.8 Flash Next memory on Apple silicon by about 27 GiB, and fixes Claude Code routing and silent truncation in /v1/responses. It follows the usual pattern of one new model plus agent-host fixes.

    View source ↗
  3. 6d ago

    v0.40.16-rc0: Ripple Preview support in MLX runner

    The candidate that added Ripple to the MLX runner with fused kernels and regression tests, folded into v0.40.16 a day later.

    View source ↗
  4. 8d ago

    v0.40.15: MiMo V2.6, AutoTrust models, Kimi Code CLI launch

    The stable cut adds MiMo V2.6 and AutoTrust decision models, extends `ollama launch` to Kimi Code CLI and Pi, and lets recurrent and vision GGUF models serve concurrent requests. It continues the pattern of pairing new models with more agent-host plumbing.

    View source ↗
  5. 8d ago

    v0.40.15-rc2: Kimi Code CLI support in launch

    A release candidate adding Kimi Code CLI to `ollama launch`, folded into the v0.40.15 stable hours later.

    View source ↗
  6. 9d ago

    v0.40.15-rc1: llama.cpp: version bump b12111 (#19530)

    A llama.cpp bump that drops a Metal patch now fixed upstream. Engine upkeep ahead of the stable release.

    View source ↗